Re: [PATCH 16/18] tty: serial: samsung: shrink the clock selection to 8 clocks

From: Tudor Ambarus
Date: Wed Jan 17 2024 - 11:27:01 EST




On 1/16/24 19:09, Sam Protsenko wrote:
> On Wed, Jan 10, 2024 at 4:25 AM Tudor Ambarus <tudor.ambarus@xxxxxxxxxx> wrote:
>>
>> <linux/serial_s3c.h> provides a clock selection pool of maximum 4 clocks.
>
> Then maybe it makes sense to turn those two field into 4-bit bit
> fields? More importantly, what particular problem does this patch
> solve, is this optimization really needed, and why? I'm not saying
> it's not needed, just that commit message might've been more verbose
> about this.
>

I guess I could have been more verbose in the phrase from below and said
that for arm64 ``struct s3c24xx_uart_info`` spans through 2 cachelines
and contains 2 holes, and with a bit of love it can fit a single
cacheline with no holes. The end goal is to reduce the memory footprint
of that struct.

I chose u8 and allowed a max of 8 clocks simple because it's large
enough to allow more clocks than are supported by the driver now, and
not too big to cause spanning of the structure through 2 cachelines.


>> Update the driver to consider a pool selection of maximum 8 clocks. The
>> final scope is to reduce the memory footprint of
>> ``struct s3c24xx_uart_info``.
>>
>> Signed-off-by: Tudor Ambarus <tudor.ambarus@xxxxxxxxxx>
>> ---
>> drivers/tty/serial/samsung_tty.c | 22 +++++++++++-----------
>> 1 file changed, 11 insertions(+), 11 deletions(-)
>>
>> diff --git a/drivers/tty/serial/samsung_tty.c b/drivers/tty/serial/samsung_tty.c
>> index 436739cf9225..5df2bcebf9fb 100644
>> --- a/drivers/tty/serial/samsung_tty.c
>> +++ b/drivers/tty/serial/samsung_tty.c
>> @@ -81,11 +81,11 @@ struct s3c24xx_uart_info {
>> unsigned long tx_fifomask;
>> unsigned long tx_fifoshift;
>> unsigned long tx_fifofull;
>> - unsigned int def_clk_sel;
>> - unsigned long num_clks;
>> unsigned long clksel_mask;
>> unsigned long clksel_shift;
>> unsigned long ucon_mask;
>> + u8 def_clk_sel;
>> + u8 num_clks;
>> u8 iotype;
>>